BIM Bench
Agents draw a timber roof eaves detail in Revit. Reviewers see the exported sheet without the model name and rate its usability from one to five.
The automated experiment compares eight models across 36 configurations on one prompt. Reviewers rate the drawing. The site does not verify native Revit geometry.
- Made by
- BIM Bench
- Kind
- CAD models
- Level
- University
- Scored
- Ranked by votes
- Answers
- Agent task
- Tasks
- All public
- Measures
- Preference
- Subject
- Civil engineering
- Language
- English
- Interaction
- Agentic
- Environment
- Autodesk Revit
- Refresh
- One fixed set
- Run by
- Anyone
- In views
- Agents
- Strongest model tested
- GPT-6 Astra #4 on the capability index
- Last model added
- October 2026
- Scores refreshed
- November 20th, 2026
GPT-6.1 Sol leads with 4
Best score per model from the benchmark's own leaderboard, read on November 20th, 2026. Where a model ran with several settings or harnesses, we keep its best. Every run is in the next chart.
Bars start at 2.4.
- 1 GPT-6.1 Sol xhigh 4
- 2 GPT-6 Astra max 4
- 3 Claude Opus 5.5 max 3
- 4
Gemini 4 Argon high 3
- 5 Claude Fable 5.1 xhigh 3
- 6
Gemini 3.8 Flash medium 3
- 7 Claude Sonnet 5.5 max 3
- 8 GPT-6.1 Terra max 3
Source: Official BIM Bench leaderboard
| # | Model | Score | Setting | Released | Run on | Link | Source |
|---|---|---|---|---|---|---|---|
| 1 | GPT-6.1 Sol | 4 | xhigh, Revit API 1.0 | Sep 2026 | Oct 2026 | bimbench.com | Official leaderboard |
| 2 | GPT-6 Astra | 4 | max, Revit API 1.0 | Sep 2026 | Oct 2026 | bimbench.com | Official leaderboard |
| 3 | Claude Opus 5.5 | 3 | max, Revit API 1.0 | Sep 2026 | Nov 2026 | bimbench.com | Official leaderboard |
| 4 | | 3 | high, Revit API 1.0 | Sep 2026 | Oct 2026 | bimbench.com | Official leaderboard |
| 5 | Claude Fable 5.1 | 3 | xhigh, Revit API 1.0 | Sep 2026 | Nov 2026 | bimbench.com | Official leaderboard |
| 6 | | 3 | medium, Revit API 1.0 | Sep 2026 | Oct 2026 | bimbench.com | Official leaderboard |
| 7 | Claude Sonnet 5.5 | 3 | max, Revit API 1.0 | Sep 2026 | Nov 2026 | bimbench.com | Official leaderboard |
| 8 | GPT-6.1 Terra | 3 | max, Revit API 1.0 | Oct 2026 | Oct 2026 | bimbench.com | Official leaderboard |
The best score, 4, costs $0.59 a task
Runs on the cost frontier score higher than any run that's cheaper. The costs are the leaderboard's own estimates, so compare them only within this benchmark.
Source: Official BIM Bench leaderboard
| Model | Effort | Harness | Score | Cost per task | Output tokens |
|---|---|---|---|---|---|
| GPT-6.1 Sol | xhigh | Revit API 1.0 | 4 | $0.59 | 29,215 |
| GPT-6 Astra | max | Revit API 1.0 | 4 | $5.04 | 47,199 |
| GPT-6 Astra | xhigh | Revit API 1.0 | 3 | $3.29 | 24,722 |
| Claude Opus 5.5 | max | Revit API 1.0 | 3 | $9.38 | 243,338 |
| GPT-6.1 Sol | max | Revit API 1.0 | 3 | $1.02 | 53,430 |
| GPT-6.1 Sol | high | Revit API 1.0 | 3 | $0.36 | 19,544 |
| Claude Opus 5.5 | high | Revit API 1.0 | 3 | $1.95 | 52,683 |
| high | Revit API 1.0 | 3 | $1.81 | 40,111 | |
| Claude Fable 5.1 | xhigh | Revit API 1.0 | 3 | $8.97 | 109,852 |
| Claude Opus 5.5 | xhigh | Revit API 1.0 | 3 | $3.83 | 99,521 |
| medium | Revit API 1.0 | 3 | $0.56 | 38,393 | |
| medium | Revit API 1.0 | 3 | $0.83 | 28,846 | |
| Claude Sonnet 5.5 | max | Revit API 1.0 | 3 | $9.72 | 316,166 |
| high | Revit API 1.0 | 3 | $0.70 | 36,662 | |
| GPT-6 Astra | high | Revit API 1.0 | 3 | $2.34 | 16,718 |
| Claude Fable 5.1 | max | Revit API 1.0 | 3 | $9.74 | 120,334 |
| Claude Fable 5.1 | high | Revit API 1.0 | 3 | $4.17 | 51,910 |
| GPT-6 Astra | medium | Revit API 1.0 | 3 | $1.57 | 10,169 |
| GPT-6.1 Terra | max | Revit API 1.0 | 3 | $0.54 | 62,208 |
| low | Revit API 1.0 | 3 | $1.99 | 62,140 | |
| GPT-6.1 Sol | medium | Revit API 1.0 | 2 | $0.23 | 7,820 |
| Claude Fable 5.1 | medium | Revit API 1.0 | 2 | $3.58 | 43,775 |
| low | Revit API 1.0 | 2 | $0.72 | 47,227 | |
| Claude Sonnet 5.5 | xhigh | Revit API 1.0 | 2 | $3.67 | 148,316 |
| Claude Opus 5.5 | medium | Revit API 1.0 | 2 | $1.16 | 31,026 |
| GPT-6.1 Terra | high | Revit API 1.0 | 2 | $0.15 | 13,392 |
| GPT-6 Astra | low | Revit API 1.0 | 2 | $0.54 | 3,367 |
| GPT-6.1 Terra | medium | Revit API 1.0 | 2 | $0.09 | 8,369 |
| GPT-6.1 Sol | low | Revit API 1.0 | 2 | $0.09 | 4,646 |
| GPT-6.1 Terra | xhigh | Revit API 1.0 | 2 | $0.27 | 28,385 |
| Claude Sonnet 5.5 | high | Revit API 1.0 | 2 | $1.16 | 55,757 |
| Claude Fable 5.1 | low | Revit API 1.0 | 1 | $2.17 | 27,295 |
| GPT-6.1 Terra | low | Revit API 1.0 | 1 | $0.04 | 2,953 |
| Claude Opus 5.5 | low | Revit API 1.0 | 1 | $0.43 | 12,535 |
| Claude Sonnet 5.5 | medium | Revit API 1.0 | 1 | $0.25 | 13,051 |
| Claude Sonnet 5.5 | low | Revit API 1.0 | 1 | $0.33 | 16,649 |
3 models have held number one on the official leaderboard since 2023
Lab claims count from the day the card came out, independent runs from the model’s release.
Source: Official BIM Bench leaderboard
Scores over time
All the scores we have, one dot each, so a model tested at several settings appears more than once.
Source: Official BIM Bench leaderboard
No lab model card reports it yet
We read the headline results of 540 launch posts from 57 labs, and none of them list this benchmark.
More in Engineering and CAD
- EEBench In use 13 electrical-engineering circuit and PCB design tasks graded by deterministic build checks, simulation and BOM constraints. No LLM judge.
Gemini 4 Carbon 83.9%
- 3DCodeBench In use Agentic procedural 3D modeling via code: author engine-ready 3D assets through software APIs. Graded for executability and shape fidelity.
LongCat-2.5 55.4%
- CADGenBench Only in system cards Agents construct CAD models from design prompts. The generation split is graded for executability and geometric correctness. Claude Fable 5 45.1%
- PartBench Only in system cards Parametric mechanical part design and CAD-style generation, graded for geometric and manufacturing fidelity. Grok 4.6 58.5%
- Razavi-Bench Not in model cards Models explain what happens in analog circuits and why. A language model awards up to four points per answer against a reference and a rubric.
Gemini 4 Carbon 98.3%
- CircuitRubric Not in model cards Models write analog circuit netlists from design requests. A program compares transistor connections and relative device sizes with accepted reference circuits. Claude Opus 4.8 96%
Data audit
All 10 checks pass. Every benchmark
Catalog record 2 pass
- Pass Task access All tasks are public.
- Pass Confidence Checked against a primary source.
Scores 4 pass 1 note
- Pass Independent runs Scores from the official leaderboard (8 entries).
- Pass Models scored 8 models we track.
- Pass Newest score The newest is from October 2026, for GPT-6.1 Terra.
- Pass Frontier models 7 of the top 10 have a score, led by GPT-6 Astra (#4).
- Note Headroom Scored as a rating, with no fixed ceiling.
Leaderboard import 3 pass
- Pass Importer We read the board on bimbench.com: 36 rows for models we track at the last read.
- Pass Last read November 20th, 2026.
- Pass Read worked The last read worked.
Consistency 1 pass 1 note
- Note Lab claims No model has both a lab number and an independent run.
- Pass Known issues No published audit found problems.